Work under mechanical engineering (Dr. Ehsan Atefi) and physics department faculty (Dr. Farrukh Fattoyev) on American Chemical Society grant project to explore novel computational models of pendant drop tensiometry. Developed framework for extracting physical features from images of pendant drops using the canny edge detector and mathematical/image processing. Used this environment to generate feature data sets and developed XGBoost and LightGBM regression models to predict Beta values of a given pendant drop profile. Compared these results to outperform traditional mathematical models such as Young-Laplace equation and modern deep learning solutions.

• Develop functional Software-Hardware co-design of an embedded RISC-V Heterogeneous SoC that minimizes end-to-end latency in real-time computer vision workloads such as YOLOv3, YOLOv7 CNNs.• Identified bottlenecks in pre-processing of images and in 30% of custom YOLOv3 layers in RTL simulation that show vectorization potential. Implemented base workload on quad-core Rocket Chip and Nvidia loosely coupled deep learning accelerator (NVDLA)• Implementing and characterizing timing and memory performance improvement of designed RISC-V SoC and vectorization workload with a tightly coupled vector accelerator on RTL simulation and FPGA prototype.

A startup focused on automotive security, ADAS, and embedded systems